This thesis describes a new path tracking algorithm, a wheel controller design and an absolute position estimation method for a free ranging mobile robot driven by two difference wheels. The smooth path tracking motion of the mobile robot with accurate velocity control of the driving wheels is very important for mobile robot naviation. To achieve this, a more improved path tracking control scheme and design of wheel velocity controller are required. Moreover accurate and reliable estimation of mobile robot position in absolute coordinate is essential to provide the path tracking controller with information about the current position of mobile robot. To complete overall control system for the mobile robot, the research presented in this paper consists of three parts: the path tracking algorithm, wheel controller design and the absolute position estimation method. 1) Path tracking algorithm A new path tracking algorithm which determines the reference wheel velocities of two driving wheels in order for the mobile robot to accurately track the desired path is proposed. The proposed scheme is designed by considering the acceleration limits of linear and rotational motions of the mobile robot in order to achieve smooth tracking motion avoiding the slippage as possible. The linear motion control alorithm is constucted based on the time optimal Bang-Bang control theory and the rotational motion control algorithm is based on the design of a landing curve satisfying the curvature constraints and the acceleration limits of the mobile robot. To prove the validity of the proposed method, a series of path tracking experiments by using dead reckoning position estimation are performed on a mobile robot (LCAR). The reuslts show that the smoothly converging motion to the various test path can be obtained with good tracking performance. 2) Wheel controller design To improve the tracking performance of the wheel velocities, an adaptive feedforward controller based on the approximate inverse model of wheel dynamics is proposed. To avoid the non-minimum phase problem, the approximate invese model is obtained by acquiring the partial sum from the infinite power series of the inverse transfer function of the plant. In the proposed scheme, the on-line parameter estimation scheme based on the least square method is utilized to cope with the uncertainties in plant parameters and variation in control circumstance. To show the effectiveness of the proposed control scheme, a series of experiments are performed. The experimental results show that the addition of the proposed feedforward control loop to feedback controller improves the velocity tracking performance of the wheels and results in more accurate and stable path tracking motion than the case of the conventional control. 3) Absolute position estimation A new position estimation method for mobile robot navigation that exploits the artificial neural network in constructing the relationship between camera and landmark coordinates is presented. To avoid the limitations associated with conventional methods such as imperfect camera calibration or finding mathematical solution which may not exist, a neural net-based approach directly relating feature points in the mark pattern to the camera location. The effectiveness of the use of a 3-D kandmark designed to increase sensitivity to image variation with camera location is discussed in some details. The experimental results show that the estimation error is within satisfactory bound irrelevant of the measuring region, type of paths and environmental variation such as crossing over a bumper or slippage. Based on these results, it is expected that the respective proposed methods can contribute to comlete vision based navigation system for mobile robot in out-door environment in near future.
